Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
V
o
l.
6, N
o
. 4
,
A
ugu
st
2016
, pp
. 18
89
~
1
896
I
S
SN
: 208
8-8
7
0
8
,
D
O
I
:
10.115
91
/ij
ece.v6
i
4.9
513
1
889
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJECE
Respond Rank: Improving Ranking
of Answers in Community
Questi
on
Answeri
ng
Geerthik S
1
, K. Rajiv
G
a
n
dhi
2
,
V
e
n
k
a
t
ra
ma
n
S
1
1
Com
puter S
c
ie
nce,
P
R
IS
T Uni
v
ers
i
t
y
,
Indi
a
2
Alagappa
Univ
ersit
y
Const
ituen
t Coll
ege
,
Ind
i
a
Article Info
A
B
STRAC
T
Article histo
r
y:
Received Nov 29, 2015
Rev
i
sed
Jun
24,
201
6
Accepte
d J
u
l
8, 2016
Ranking is used in Community
Question Answe
r
ing (CQA) for
positioning
user answers. Different rank
ing techni
qu
es ar
e us
ed in CQA for
r
a
nking user
ans
w
ers
.
W
e
id
entifi
e
d thr
ee d
r
awb
acks with
the ex
isting ran
k
ing. Th
e
Qualit
y
answers
written recen
tl
y
are not recogni
zed properl
y
com
p
ared to old
averag
e ans
w
ers
.
Als
o
the ans
w
ers
from
the us
ers
having les
s
num
ber of
followers ar
e no
t recognized pro
p
erly
compared
to users with m
o
re number
of followers
. Mo
reover
exper
t
s and norma
l us
er
l
i
kes
ar
e
trea
ted
equal
l
y
. W
e
propose RespondRank for ranking the user
ans
w
ers. RespondRank iden
tifies
be
st a
n
swe
r
bette
r
c
o
mpa
r
e
d
to existing me
thods.With Re
spondRa
nk Qua
lity
answers from different us
ers are rec
ognized pro
p
erly
.
Exper
i
ments carried
out on Quora a popular CQA, shows our
Resp
ondRank shows significan
t
improvement
in
ranking than
the existing
ranking
techn
i
ques.
Keyword:
CQA
R
a
nki
ng
t
ech
ni
que
s
Resp
ond
Ran
k
Copyright ©
201
6 Institut
e
o
f
Ad
vanced
Engin
eer
ing and S
c
i
e
nce.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
Geert
h
ik S,
Research Sc
holar Com
puter
Science,
PRIST Un
iv
ersity,
In
dia.
Em
a
il:
1.
INTRODUCTION
C
o
m
m
uni
ty
quest
i
on answeri
ng (C
Q
A
) i
s
a t
y
pe of i
n
for
m
ati
on ret
r
i
e
val
were user ne
ed from
t
h
e
com
m
uni
ty
i
s
gi
ven i
n
t
h
e for
m
of nat
u
ral
l
a
ng
uage que
st
i
o
n and com
m
uni
t
y
response i
s
i
n
t
h
e form
of nat
u
ral
language a
n
swer. The
great success of CQA leads to pr
ogram
m
i
ng wizards like stackov
erflow.c
om
[1], Quora
a place for
knowledge sha
r
ing [2
],[3], yahoo answers
[4], z
h
ihu
[5] etc.
Many users re
gister with
CQA
and
onl
y
a few are
act
i
v
e. Anal
y
z
ing a Ja
va f
o
ru
m
[6]
wi
t
h
soci
al
net
w
ork an
al
y
s
i
s
m
e
t
hod,
t
h
e resul
t
s
sho
w
t
h
a
t
onl
y
12 % o
f
u
s
ers ask quest
i
ons an
d ans
w
er t
h
e quest
i
ons
am
ong t
h
em
selves. They
are o
n
l
y
t
h
e hi
ghl
y
act
i
v
e
user
.
Al
so o
n
l
y
13 % of users onl
y
answer q
u
est
i
ons an
d near
l
y
hal
f
of the users o
n
l
y
ask
questions.
Quora
contain
nearly 100 m
i
lli
on users we
re 80
% of people are
only visitors
who read the a
n
swer
cont
ent
.
O
n
l
y
20 %
o
f
users
vot
e t
h
e a
n
swe
r
an
d f
o
l
l
o
w
pe
opl
e. I
n
m
a
jori
t
y
of
C
QA'
s ra
nki
n
g
i
s
base
d
on
onl
y
a sm
all
crowd
of
users.
New challenge
s for researc
h
e
r
s are created by CQA
sites. So
m
e
o
f
th
e
ch
allen
g
e
s we id
en
tified
in
C
QA sy
st
em
s
are i
d
ent
i
f
i
cation
of e
xpert
s,
answer
ranki
ng
, deal
i
ng wi
t
h
unf
orm
a
tt
ed qu
est
i
ons, deal
i
ng wi
t
h
unans
wered
qu
est
i
ons. The users who are w
r
i
t
i
ng t
h
e answer t
o
t
h
e quest
i
ons
m
a
y
be expert
i
n
t
h
e subject
he i
s
an
swering
or
n
e
w to
th
e
p
a
rticu
l
ar to
p
i
c. As a resu
lt, mo
st o
f
th
e C
Q
A sites fail in
th
e q
u
a
lity ran
k
i
n
g
o
f
answers. S
o
t
h
e
m
o
st
im
portant
chal
l
e
nge in C
QA i
s
di
scoveri
n
g
a qual
i
t
y
answer fr
o
m
group
of an
swers
written
b
y
d
i
fferen
t
u
s
ers
[7
].
Co
n
s
id
erin
g
the q
u
e
stio
n
s
, th
ey are p
o
s
ted
b
y
u
s
ers in
d
i
fferen
t
d
o
m
a
i
n
s
. Th
e q
u
a
lity o
f
th
e q
u
e
stio
n
can be m
e
asured wi
t
h
t
h
e n
u
m
b
er of
up
vo
t
e
s
and
d
o
w
n
v
o
t
e
s
for t
h
e
qu
est
i
on, quest
i
on t
a
g, l
e
ngt
h
of t
h
e
question and
the num
ber o
f
answers received to the
question [8]. Different
ty
pes
of questions
posted
i
n
C
QA
are de
fi
ni
t
i
on, opi
ni
o
n
, pr
ocedu
r
e, r
eason, fact
oi
d,
Why
,
Yes/
N
o
q
u
est
i
ons. Am
o
ng t
h
ese
fact
oi
d and
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE
Vo
l. 6
,
N
o
. 4
,
Au
gu
st 2
016
:
18
89
–
1
896
1
890
Yes/No questions are difficul
t
to answ
er because they need
m
o
re fac
t
s and
explanation in answering [9]. Also,
t
h
e fact
oi
d, op
i
n
i
on and
defi
ni
t
i
on quest
i
o
n
s
were at
t
r
ac
t
i
n
g m
o
re users to wri
t
e
answer
s t
h
an ot
her t
y
pes of
quest
i
ons [
10]
.
Fact
oi
d answers are answ
ere
d
wi
t
h
t
h
e hel
p
of o
p
i
n
i
ons w
h
i
c
h i
s
based o
n
pers
ons desi
r
e
,bel
i
e
f
or spec
ul
at
i
ons [11]
.
Al
so
Why
q
u
est
i
o
n
i
s
expl
ai
ned by
user o
n
l
y
wi
t
h
t
h
e hel
p
of m
a
teri
al
,person
or a
pur
pose [1
2]
.
Co
n
s
id
erin
g
the an
swers, th
e an
swer
written
to
th
e q
u
e
stio
n
m
u
st
b
e
rele
v
a
n
t
to
th
e q
u
e
stio
n
.
So
m
e
q
u
a
lity an
swers h
a
v
e
an in
trod
u
c
to
ry
p
a
rt,
p
l
o
t
an
d th
em
e o
f
an
swer. Th
e
u
s
er m
u
st b
e
patien
ce in
writi
n
g
the
answer and if needed the a
n
swer s
h
ould
be descriptive. Som
e
answers
need som
e
proofs and also some
im
ages. So the
best answer must att
ach
th
e imag
es o
r
th
e URL lin
k
fo
r th
e
refere
nces if
needed. Som
e
question
onl
y
needs a
fact
ual
answer a
nd
for
such
t
y
pe of
quest
i
ons t
h
e answe
r
m
u
st
be sho
r
t
and c
l
ear. C
onsi
d
eri
ng t
h
e
users the best users in
CQA
have
three properties. They
m
u
st give
tim
ely response, they provide
quality
answers, a
nd t
h
ey
m
a
i
n
t
a
i
n
a goo
d soci
al
pro
f
i
l
e
[13]
.
Thi
s
paper i
s
o
r
gani
zed as f
o
l
l
o
ws:
sect
i
on 2 di
scusses t
h
e pro
b
l
e
m
s
wi
t
h
t
h
e current
ra
n
k
i
ng sy
st
em
,
sect
i
on 3 descr
i
be about
t
h
e r
e
l
a
t
e
d work i
n
ranki
n
g
C
Q
A,
sect
i
on 4 descr
i
bes about
R
e
s
p
o
ndR
an
k
al
go
ri
t
h
m
i
n
ran
k
i
ng,
sect
i
on 5
gi
ves t
h
e resul
t
s
and
d
i
scussi
on,
t
h
e f
u
t
u
re
wor
k
a
n
d m
a
i
n
concl
u
si
ons are
gi
ven
i
n
t
h
e
last sec
tion.
2.
PROBLEMS WITH RANKING ANSWE
R
S
CQA
sites are creating arc
h
i
v
es of hundre
d
s of qu
est
i
o
n
s
and m
i
ll
i
ons
of ans
w
e
r
s every
day
.
S
o
id
en
tifying
th
e b
e
st an
swer is a b
a
sic n
e
ed
in
CQA
site
s. Diffe
re
nt facto
r
s use
d
by
CQ
A
sites f
o
r
ran
k
in
g th
e
an
swers are, t
o
tal
up
vo
tes
t
o
th
e an
swer, to
tal
dow
nvo
tes
to
th
e an
swer, prev
io
u
s
answers
written
b
y
th
e
au
tho
r
, wh
eth
e
r th
e au
t
h
or is an
exp
e
rt in
t
h
e su
bj
ect or not an
d
qu
ality o
f
th
e con
t
en
t, etc.
W
ith
th
e facto
r
s
m
e
nt
i
oned a
b
o
v
e t
h
e a
n
swe
r
s
wi
t
h
m
o
re
u
p
v
o
t
e
s
are
ra
nke
d
hi
g
h
er a
nd t
h
e
answe
r
s
wi
t
h
m
o
re
do
w
n
v
o
t
e
s
are
ran
k
e
d
lo
wer
.
In
an
y
CQA
si
te p
e
op
le will b
e
read
ing
on
l
y
t
o
p
two
o
r
t
h
ree an
swers
an
d
upv
o
t
e
an
y
o
n
e
o
f
the
an
swers. Th
e an
swers
wh
ich
are no
t
in
itiall
y written
and
in
th
e bo
tto
m
a
r
e no
t read
and co
n
s
i
d
ered
b
y
m
a
n
y
u
s
er
s. On
e m
e
t
h
od
in
r
a
nk
i
n
g to
ov
er
co
m
e
t
h
is pr
ob
lem
is
to
h
i
d
e
t
h
e likes for th
e
u
s
ers for all an
swers fo
r
som
e
peri
od o
f
t
i
m
e
. If a user
answe
r
got
ve
ry
go
od
u
p
vot
e
com
p
ared t
o
o
t
her use
r
ans
w
ers, t
h
e
n
t
h
e answe
r
is declared as t
h
e best answe
r
and the
n
upvotes
are vi
sib
l
e
to
all u
s
ers
.
B
u
t p
r
actical i
m
p
l
em
en
tatio
n
o
f
th
is
m
e
t
hod i
s
di
ffi
cul
t
.
A basi
c st
at
i
s
t
i
cal
m
e
t
hod o
f
ran
k
i
n
g i
s
, i
d
e
n
t
i
f
y
i
ng t
h
e
us
ers w
ho a
n
s
w
e
r
m
o
re on a pa
rt
i
c
ul
ar t
o
pi
c
and
co
nsi
d
e
r
h
i
m
as an ex
per
t
and
gi
ve
hi
s
up
v
o
t
e
m
o
re
im
p
o
r
tan
ce. B
u
t th
is is n
o
t
t
r
ue, in
th
e case
of on
lin
e
adve
rt
i
s
em
ent
and
s
p
am
m
e
rs wh
o a
r
e i
n
t
e
n
d
e
d i
n
p
r
om
ot
i
ng t
h
ei
r
pr
od
uct
s
o
n
l
i
n
e.
The re
st
of t
h
i
s
sect
i
on
gi
ves t
h
e bac
k
gr
o
u
n
d
anal
y
s
i
s
of
fol
l
owe
r
s i
n
C
Q
A
,rel
at
i
o
nshi
p
b
e
t
w
een t
h
e
num
ber o
f
f
o
l
l
o
wers a
n
d answe
r
vi
e
w
a
nd al
s
o
di
sc
u
ss t
h
e t
w
o
m
a
jor
pr
o
b
l
e
m
s
we i
d
ent
i
f
i
e
d i
n
the CQA syste
m
.
2.
1.
Back
ground
analysis
on foll
ower in
CQ
A
Fo
llo
wers
o
n
C
QA
fo
r an
y u
s
er d
e
p
e
n
d
s
o
n
th
e qu
ality o
f
h
i
s an
swer, b
l
og
s h
e
m
a
in
tain
s, wh
eth
e
r
h
e
i
s
a celebri
t
y
o
r
not
, t
o
pi
c of e
xpert
i
s
e, gen
d
e
r
, et
c. At
t
h
e sam
e
tim
e, t
h
e num
ber of fol
l
o
wers al
so depe
nds o
n
t
h
e t
o
t
a
l
t
i
m
e
a user i
s
spendi
ng o
n
C
Q
A
si
t
e
s. C
onsi
d
er a user Al
ex K. C
h
en w
ho aske
d
45,
11
3 quest
i
ons i
n
Quo
r
a. He had
m
o
re
t
h
an 12,0
00 f
o
l
l
o
wers.
C
onsi
d
er
anot
h
e
r user M
a
rc Bod
n
i
c
k wh
o asked 1
2
,5
91 q
u
e
s
t
i
ons
and he had m
o
re t
h
an 61,
39
1 fol
l
o
wers. O
n
t
h
e ot
her ha
n
d
, 60% of
users i
n
Q
uora
had l
e
ss t
h
an 10 fol
l
o
wers
.
If a n
e
w u
s
er reg
i
ster to
CQA
sites, th
e n
u
m
b
er o
f
fo
llo
wers, h
e
go
t will b
e
v
e
ry less.
2.2.
Relations
h
ip
between num
b
er of
followe
rs and
answer
view
The t
o
t
a
l
nu
m
b
er of
up
vot
es
al
so depends
on t
o
t
a
l
nu
m
b
er of answer v
i
ews.
Answer
vi
ew i
s
t
h
e
opt
i
o
n
by
w
h
i
c
h use
r
can
read
one
o
r
t
w
o l
i
n
es i
n
every
a
n
s
w
er. If he
is
int
e
rested
in readi
ng t
h
e entire a
n
swer,
h
e
can
v
i
ew the en
tire an
swer. An
swer
v
i
ew is
m
o
re
fo
r the an
swers
wh
i
c
h
are written
b
y
u
s
ers wh
o co
n
t
ain
man
y
fo
llo
wers. Co
n
s
id
er a user Jo
el, wh
o had
written
an
an
swer for a
q
u
estio
n
.
Co
n
s
id
er if ano
t
h
e
r
u
s
er Sco
t
t
who follow J
o
el,
upvotes
Joel answer.
No
w
th
e Jo
el an
swer is v
i
sib
l
e to
Sco
tt fo
llo
wers. If th
e fo
llo
wers of
Sco
tt
lik
es Jo
el an
swer
th
ey
up
vo
te
it. If Scott is h
a
ving m
a
ny followe
rs
t
h
e ans
w
er
views is eve
n
increased.
It
i
s
cl
ear t
h
at
there i
s
a st
ron
g
rel
a
t
i
onshi
p be
t
w
een t
h
e num
ber o
f
f
o
l
l
o
wer
s
and
num
ber o
f
answe
r
vi
ews
.
2.3.
Problem 1:
Pr
oblems faced by new
users
In m
o
st of th
e
CQA
sites like
Quora,
Stackove
rflow
any
user can contri
bute the answer to the
quest
i
on. B
u
t
m
o
st
of t
h
i
s
si
te i
s
very
unfai
r
t
o
t
h
e new
users and t
h
e a
n
s
w
ers p
r
o
v
i
d
ed
by
t
h
em
. The quest
i
o
n
and answe
r
s users read in their tim
el
i
n
e ar
e from
t
h
e peopl
e and st
uff t
h
ey
fol
l
o
w [5]
,
[1
3]
. The ranki
ng o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Resp
on
d
Ra
nk:
Im
pr
ovi
n
g
Ra
nki
n
g
of
A
n
sw
e
rs i
n
C
o
m
m
uni
t
y
Que
s
t
i
o
n
An
sw
eri
n
g
(
G
eert
h
i
k
S)
1
891
an
swers is
m
o
stly
d
e
ter
m
in
ed
b
y
th
e n
u
m
b
e
r o
f
fo
llo
wers
for the user
who
write
s the answer and not th
e
q
u
a
lity o
f
th
e an
swer.
So
m
o
st o
f
th
e an
swers rank
ed
top
s
are fro
m
th
e u
s
ers who
h
a
v
e
m
o
re fo
llo
wers ev
en
t
hou
gh t
h
e ans
w
ers m
a
y
not
be i
n
t
e
resti
ng so
m
e
tim
es. In ot
her w
o
rds
,
i
f
a new user
wi
t
h
few fol
l
o
we
rs,
wri
t
e
s
a good ans
w
er, his answer is not ranke
d in top. The
go
od answers are not reached to
the maxim
u
m users due to
th
is p
r
ob
le
m
.
Topi
c cat
egori
zat
i
on [14]
i
s
f
o
u
nd i
n
m
o
st
of t
h
e
C
QA
si
t
e
s whi
c
h hel
p
users
m
a
i
n
ly
to reduce t
h
e
ti
me in
search
i
n
g
an
swers. B
u
t if
q
u
a
lity a
n
swers are
n
o
t
ran
k
e
d
in
to
p u
s
ers n
e
ed
to sp
en
d m
o
re ti
me in
readi
ng al
l
t
h
e answers f
o
r t
h
at
part
i
c
ul
ar quest
i
on f
o
r g
e
t
t
i
ng t
h
e need
ed kn
owl
e
d
g
e.
M
a
ny
users wri
t
e
t
h
e
answer i
n
C
Q
A
si
t
e
s for s
o
m
e
personal
h
a
ppi
ness t
h
ey
got
fr
om
t
h
e
upv
ot
es.
So m
a
ny
new
users
were n
o
t
willin
g
to
write an
swers in th
e
CQA sites.
2.
4.
Problem 2:
Pr
oblem face
d
by
L
a
te Quality answers
To
illu
strate th
i
s
p
r
ob
lem
,
co
n
s
id
er th
e Tab
l
e 1
,
wh
ere a
u
s
er ask
s
a qu
estio
n
in
th
e b
i
o-m
e
d
i
cal field
.
Tabl
e 1. Wee
k
vrs
R
a
n
k
User
s
W
e
ek
Rank
obtained
UserA1
W
e
ek 1
Rank 1
User A2
W
e
ek 1
Rank 2
- -
-
User A50
W
e
ek 1
Rank 50
User A51
W
e
ek 2
Rank 51
Let'
s consi
d
er if
UserA1
answ
er t
h
e quest
i
on i
n
t
h
e f
i
rst
week i
t
s
el
f and i
t
i
s
vot
ed by
m
a
ny
users as
t
h
e best
answe
r
fr
om
a l
i
s
t
of
50 a
n
swers
.
C
o
nsi
d
er i
f
UserA
5
1
a
n
swers
the
sa
m
e
ques
tion after one
week and
the answer is
better co
m
p
ared to
UserA
1
an
swer.
Sin
ce th
e an
swer is wr
itten
after o
n
e
week
in
itially
th
e
answer i
s
di
sp
l
a
y
e
d i
n
51 po
si
t
i
on.
In m
o
st of t
h
e
C
Q
A
si
t
e
s
m
a
ny
users read onl
y
t
o
p
few answe
r
s.
Aft
e
r
readi
ng t
w
o
or
t
h
ree t
op answers, t
h
ey
vot
e any
one of t
h
e top answers
lis
ted at
that
tim
e. So
in
m
o
st of the
cases, an
swer
written
b
y
Use
r
A51
i
s
not
re
ad by
m
a
ny
users. Eve
n
t
h
ou
gh
UserA5
1
answer is m
o
re quality
answer than the
UserA1
ans
w
er i
t
i
s
not
v
i
si
bl
e
t
o
all
users. In o
u
r
R
e
s
p
o
ndR
an
k
m
e
tho
d
t
h
i
s
probl
em
i
s
rem
oved by
usi
ng a
n
swer
vi
ew as o
n
e o
f
t
h
e
feat
ure i
n
t
h
e r
a
nki
ng
sy
st
em
.
B
y
usi
ng t
h
e p
e
rcent
a
ge
of a
n
swer vi
ews a
n
d perce
n
t
a
ge o
f
fol
l
o
we
r
u
p
v
o
t
e
i
n
ou
r ran
k
i
ng m
e
t
hod
t
h
e pro
b
l
e
m
s
gi
ven abo
v
e are
great
l
y
reduced
by
R
e
spon
dR
ank al
g
o
ri
t
h
m
.
3.
RELATED WORK
A h
ybrid
h
i
erarch
y-o
f
-classi
fiers fram
e
wo
rk
fo
r
fi
n
d
i
ng
t
h
e qu
ality an
s
w
ers in
yaho
o an
swers is
pro
pose
d
by
[9
]
.
B
e
fore anal
yzi
ng t
h
e di
ffere
nt
answers t
o
a gi
ven quest
i
o
n
,
t
h
e quest
i
on i
s
anal
y
zed fi
rst. The
user answers are co
m
p
ared w
ith the expected answ
ers which are already stor
ed fo
r t
h
e di
f
f
erent
quest
i
on t
y
pes
.
The fram
e
work i
s
com
p
ared wi
t
h
di
fferent
quest
i
ons f
r
om
y
a
hoo ans
w
er
s and t
h
e best
answer
predi
c
t
i
on i
s
very
goo
d. Usa
g
e of
z-sco
r
e m
easures i
n
i
d
ent
i
f
y
i
ng t
h
e expert
s i
s
done i
n
[6]
were i
f
the users ask an
d repl
y
equal
num
ber of t
i
m
e
s on a part
i
c
ul
ar t
opi
c hi
s z-score i
s
0. If
t
h
e user
answer m
o
re t
h
an t
h
e
quest
i
ons
he
p
o
s
ted
fo
r a p
a
rticu
l
ar to
p
i
c h
i
s z-sco
r
e is p
o
sitiv
e v
a
lu
e e
l
se
if th
e u
s
er
ask
m
o
re o
n
a p
a
rticu
l
ar to
p
i
c an
d
answer
fewer
h
i
s z-score i
s
ne
gat
i
v
e. The p
r
o
b
l
e
m
wi
t
h
z-score al
go
ri
t
h
m
is t
h
i
s
al
gori
t
h
m
onl
y
based
on t
h
e
num
ber of re
pl
i
e
s and n
o
t
bas
e
d o
n
g
o
o
d
re
p
l
i
e
s on part
i
c
ul
ar t
opi
cs.
An a
n
alogical reasoning approac
h
fo
r r
a
nk
in
g
,
wh
er
e
ran
k
i
n
g
fo
r th
e an
sw
er
s is based
on
the
resem
b
l
a
nce bet
w
een new
que
st
i
on answers a
nd exi
s
t
i
ng
be
st sim
i
lar question ans
w
ers
[15]. If a user writes an
answer
for
t
h
e
gi
ven
quest
i
on,
t
h
e quest
i
on a
nd a
n
swers a
r
e com
p
ared wi
t
h
t
h
e exi
s
t
i
ng
best
quest
i
on a
n
swer
of si
m
i
l
a
r ty
pe and ra
nki
n
g
i
s
do
ne. Thi
s
m
e
tho
d
i
s
go
od
fo
r
fi
ndi
n
g
t
h
e t
e
x
t
ual
m
i
sm
at
ches and a
n
swers s
p
am
.
M
o
st
of t
h
e
C
QA
uses page
rank al
go
ri
t
h
m
for ranki
n
g
answers. C
o
nsi
d
er t
h
ree users
X, Y an
d Z
.
W
ith
th
is p
a
g
e
ran
k
alg
o
rith
m
,
if th
e X an
swer to
Y an
d
Y an
swers to
Z th
en
X is co
n
s
id
ered
an
ex
p
e
rt am
o
n
g
X, Y and Z. T
h
i
s
i
s
an effi
cient
m
e
t
hod fo
r
predi
c
t
i
ng t
h
e expert
s i
n
C
QA
si
t
e
s.In l
i
k
e
m
a
nner [13]
us
es AA
page ra
nk
al
gori
t
h
m
wit
h
Q
uora
dat
a
set
s
and
cal
cul
a
t
e
d aut
h
ori
t
y
score an
d act
i
v
i
t
y
score of
u
s
ers i
n
i
d
ent
i
f
y
i
ng pot
ent
i
a
l
answer suppl
i
e
r. T
h
e
aut
hori
t
y
score
i
n
creases wi
t
h
repl
y
count
a
nd i
t
decrease
s
wi
t
h
quest
i
on cou
n
t
. Act
i
v
it
y score i
s
cal
culat
e
d based on
t
h
e frequenc
y
of user vi
si
t
i
ng
C
QA.
Her
e
t
h
e
user
u
p
v
o
t
e
wi
t
h
hi
gh aut
h
ori
t
y
and act
i
v
it
y
score get
m
o
re im
port
a
nce. Accordi
n
gl
y
Zhi
hura
nk
[5]
i
d
ent
i
f
i
e
d
t
h
e user a
u
t
h
o
r
i
t
y
i
n
ranki
ng
based
on
l
i
nk
st
ruct
ure an
d t
opi
c si
m
i
lari
ty bet
w
een
quest
i
on an
d
hi
s ex
pert
i
s
e
fi
el
d. In t
h
i
s
m
e
t
hod i
f
User X
wi
t
h
hi
gh
aut
h
o
r
i
t
y
and Use
r
Y wi
t
h
m
e
dium
aut
hori
t
y
up
vot
e
Z,
th
en
up
vo
te from u
s
er X is treated
m
o
re p
o
w
erfu
l th
an user Y
which
increa
se
aut
hori
t
y
of
user Z.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE
Vo
l. 6
,
N
o
. 4
,
Au
gu
st 2
016
:
18
89
–
1
896
1
892
4.
RESPO
N
DR
AN
K ALGO
R
I
THM
We can
ap
p
l
y
th
e
Respon
d
R
an
k
algo
rith
m
in
an
sw
er
rank
i
n
g,
b
e
st an
swer fi
n
d
i
n
g
, an
d
p
o
s
ition
i
ng
of a
n
swe
r
s.
T
h
e Fi
gu
re 1 gi
v
e
s t
h
e ove
ral
l
archi
t
ect
u
r
e of
C
QA
si
t
e
s. He
re, a use
r
p
o
st
s
a quest
i
o
n an
d ot
he
r
u
s
er
s answ
er th
e
q
u
e
stion. Fo
r
r
a
nk
ing d
i
ff
er
en
t
answers
and
positioning the
m
in an
order,
we
use Res
p
ondR
ank algorithm
.
The
best
an
swers are listed
in th
e to
p, fo
llo
wed by a
v
era
g
e
answers
followed by
i
rrel
e
va
nt
a
n
sw
ers.
Fi
gu
re 1.
A
r
chi
t
ect
ure of ra
nki
ng
i
n
C
Q
A
To
illu
strate R
e
spo
n
d
R
ank
al
g
o
rith
m
,
co
n
s
i
d
er Tab
l
e
2
.
Tabl
e 2. U
p
v
o
t
e
A
n
al
y
s
i
s
Person Ja
m
e
s
Justine
T
o
tal follower
s
2000
500
Upvotes fr
o
m
follower
s
500
180
Upvotes other
than follower
s
200
140
Answer
views fr
o
m
norm
a
l user
s
250
160
Let's assu
m
e
a
q
u
e
stion
is po
sted
an
d
it is an
sw
e
r
e
d
by
two
people Ja
m
e
s and Justine.
Assum
e
p
e
r
s
on
Jam
e
s has 200
0 fo
llo
wer
s
an
d p
e
r
s
o
n
Ju
stin
e
h
a
s 500 fo
llo
w
e
r
s
.
Assu
m
e
th
e an
swer
b
y
Jam
e
s go
t
7
5
3
upv
o
t
es
we
re 500
u
pvo
tes
are g
e
ttin
g
fro
m
th
e fo
llowers,
20
0
u
p
v
o
t
e
s
a
r
e get
t
i
ng fr
om
norm
a
l
users
and t
h
ree
e
x
p
e
rt
vot
es.
Al
s
o
, co
nsi
d
er an
swer
by
Just
i
n
e g
o
t
3
2
4
up
vo
tes were
1
8
0
up
vo
tes
are
g
e
tting
fro
m
h
i
s
fo
llo
we
rs an
d 140
v
o
t
es are g
e
tting
fro
m
o
t
h
e
r
p
e
op
le and
Just
i
n
e
got
f
o
ur e
xpe
rt
u
p
vot
es
.
W
i
t
h
Ja
m
e
s and Ju
st
i
n
e ha
vi
n
g
t
o
t
a
l
up v
o
t
e
s 7
5
3
an
d 3
2
4
u
p
vot
es
resp
ectiv
ely,
m
o
st o
f
th
e
C
Q
A
sites ran
k
th
e an
swer
p
r
ovi
ded
by
Ja
m
e
s as hi
g
h
es
t
ran
k
a
nd t
h
e
ans
w
er
p
r
ov
id
ed
b
y
Justin
e is rank
ed b
e
low t
h
e James.In
t
h
e
a
b
o
v
e
Ta
bl
e 2, pe
rcent
a
ges of
u
p
v
o
t
e
s
i
s
cal
c
u
l
a
t
e
d
a
s
gi
ve
n.
P
e
r
c
entage
ofup
v
o
tefr
o
m
f
o
llow
e
rs
Up
v
otesfr
o
mf
o
llow
ers
T
o
talnumberoff
o
llo
w
ers
100
Which is
the
best
Engineering Coll
ege in
Ca
lcutta
User Answ
ers the
question
Answ
er 1
Answ
er 2
Answ
er 3
Answ
er N
RespondRank
Algorit
hm
Top Best Answ
ers
Best Answ
er
Avera
g
e Answ
ers
Irrelevant Answ
ers
Ra
n
k
s a
n
d
o
r
d
ers th
e An
sw
ers
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Resp
on
d
Ra
nk:
Im
pr
ovi
n
g
Ra
nki
n
g
of
A
n
sw
e
rs i
n
C
o
m
m
uni
t
y
Que
s
t
i
o
n
An
sw
eri
n
g
(
G
eert
h
i
k
S)
1
893
From
t
a
bl
e 2 t
h
e pe
rce
n
t
a
ge
of
fol
l
o
we
r u
p
vot
es
of Jam
e
s and J
u
st
i
n
e a
r
e 25
per
cent
a
nd
3
6
pe
rce
n
t
resp
ectiv
ely. It
is clear th
at J
u
stin
e go
t 11
p
e
rcen
t
hi
g
h
e
r
t
h
an Jam
e
s.
C
onsi
d
er
i
ng the above scena
r
io we
un
de
rst
o
od
t
h
a
t
m
o
st
of t
h
e u
p
v
o
t
e
s
got
fr
o
m
Jam
e
s ar
e fr
om
his follo
we
rs.
Also
, if
the
r
e are m
o
re followe
rs
o
f
Justin
e th
an Jam
e
s th
en
,
su
rely Ju
stin
e
will g
e
t m
o
re upv
o
t
es th
an
the
Jam
e
s. Also
, t
h
e
n
u
m
b
e
r
of
u
pvo
tes
Just
i
n
e
got
fr
o
m
norm
a
l
user
s i
s
hi
g
h
er c
o
m
p
are
d
wi
t
h
Ja
mes. Th
e
exp
e
rt
u
pvo
tes of Ju
stin
e go
t is also
h
i
gh
er
com
p
ared
wi
t
h
Jam
e
s.B
u
t
Jus
t
i
n
e ans
w
er
ha
s ra
nke
d l
o
we
r
t
h
an
Jam
e
s wi
t
h
exi
s
t
i
n
g
ran
k
i
n
g m
e
t
hods
.
Let
us
cal
cul
a
t
e
perce
n
t
a
ge
o
f
a
n
swe
r
vi
ew
o
f
Jam
e
s an
d J
u
st
i
n
e
f
r
om
Tabl
e 2
.
P
e
r
c
entage
ofans
w
er
view
T
otalUp
v
ot
esfr
o
mnormalusers
T
o
talnum
b
erofans
w
er
view
s
100
Jam
e
s had got
20
0 up
vot
es fr
om
250 ans
w
er
vi
ews and Just
i
n
e had got
1
4
0
up
vot
es fr
om
160 ans
w
er
vi
ews. A
ppl
y
i
ng ab
ove
for
m
u
l
a
percent
age of
answ
e
r
vi
ew
of Jam
e
s and Just
i
n
e i
s
80
percent
and
87.
5
p
e
rcen
tag
e
resp
ectiv
ely.
Mo
s
t
o
f
th
e ex
istin
g
rank
in
g m
e
t
h
od
s on
ly co
un
t th
e to
tal u
p
v
o
t
es wh
ich
mak
e
s th
e
ranki
n
g
fl
awe
d
. I
n
o
u
r R
e
s
p
o
ndR
an
k ran
k
i
ng i
s
based
on
percent
age
s of
f
o
l
l
o
w
e
r
upvot
es
and also the
percent
age
of
answ
e
r vi
ew
whi
c
h m
a
kes ranki
ng
bet
t
e
r.
4.
1.
R
e
s
p
o
n
d Ra
nk
d
e
f
i
n
i
t
i
o
n
The RespondR
ank is calculate
d as follows
Re
s
p
o
n
d
R
a
n
k
U
100
100
p
is
th
e t
o
tal nu
m
b
er
of
u
pvotes fr
o
m
th
e follo
w
e
r
s
q
- to
tal nu
m
b
er
o
f
fo
llo
wers..
r
- t
o
t
a
l
num
ber
o
f
up
v
o
t
e
s f
r
o
m
norm
a
l
users ot
her t
h
an
f
o
l
l
owe
r
s .
s
- t
o
t
a
l
a
n
s
w
er
vi
ews
f
r
om
no
r
m
al
users
ot
he
r
t
h
an
f
o
l
l
o
wers
.
α
-
nu
m
b
er
of
ex
p
e
r
t
u
pvo
tes
β
- num
ber of
e
xpe
rt
do
w
n
v
o
t
e
s
.
γ
- sp
ecial upv
otes, th
at
is t
h
e
n
u
m
b
e
r of upvo
tes
go
t
fro
m
oth
e
r u
s
ers
w
ho
alread
y
written
answ
er to th
e
same
que
st
i
o
n
.
d-
The
t
o
t
a
l
nu
m
b
er of
d
o
w
n
v
o
t
e
s f
r
o
m
nor
m
a
l
user
.
Whe
r
e
U
is t
h
e u
s
er
who
writ
es th
e an
swer
,
The
num
b
er of expert
upvote
s is m
a
rked as
α
, t
o
tal nu
m
b
er
o
f
dow
nvo
tes fr
o
m
ex
p
e
r
t
s is
d
e
no
ted
by
β
. M
o
st
of
C
Q
A
have
s
o
m
e
pot
e
n
t
i
a
l
u
s
ers
w
h
o
a
r
e
m
a
rked
as
ex
p
e
rt
s.
The
e
xpe
rt
s a
r
e i
d
ent
i
f
i
e
d
wi
t
h
s
o
me
p
a
r
a
me
te
r
s
lik
e u
s
er
s wh
o wro
t
e
th
e an
s
w
er
in
s
a
me
to
p
i
c a
l
r
e
ad
y
a
n
d
g
o
t
mo
r
e
u
pvo
te
s
.
Con
s
id
e
r
a
scenari
o
w
h
ere
si
x
use
r
s a
n
s
w
ers
t
o
a
gi
ve
n
q
u
est
i
o
n
i
n
t
h
e
fi
el
d
of
c
o
m
put
i
ng.
Am
ong
t
h
em
fo
ur
answ
er
s
appea
r
t
o
be c
o
r
r
ect
f
o
r
any
user
w
h
o
read
t
h
e an
swe
r
s.
T
h
e
best
a
n
swe
r
am
ong t
h
e
f
o
ur
i
s
u
n
d
erst
oo
d
onl
y
by an expe
rt in com
puting.
In
som
e
cases t
h
e
user
a
n
swe
r
a
quest
i
o
n
an
d
he
g
o
t
s
o
m
e
upv
ot
es a
n
d i
f
h
e
fi
n
d
s
som
e
ot
her
ans
w
e
r
is m
o
re in
terestin
g th
an
h
i
s an
swer or if h
e
lik
es so
m
e
o
t
h
e
r an
swer h
e
will b
e
ab
le t
o
upv
o
t
e th
at
an
swer
.
That
upvotes a
r
e s
p
ecial upvotes and they a
r
e de
note
d
by
γ
.
It is a
v
e
ry
rare situ
atio
n th
e
an
swer
g
e
ts the do
wn
v
o
t
es.
Bu
t if th
e an
swer is to
tally irrelev
a
n
t
to the
current
que
stion, s
o
m
e
user will downvote the answe
r
.
In our m
e
thod we took t
h
e val
u
e of upvote, downvote
,exp
er
t
up
vo
te
an
d exp
e
r
t
dow
nvo
te as
on
e.
Bu
t
in
real-ti
m
e im
ple
m
entation,
we can increase the
expe
rt
u
pvo
te and
downvo
te v
a
l
u
e
based
on
th
e quality o
f
exp
e
rt.
4.
2.
C
o
mp
ut
in
g Res
p
o
n
d
R
a
nk
C
onsi
d
er
T
a
bl
e
3
w
h
ere
det
a
i
l
s
o
f
2
users
are
gi
ve
n.
T
h
e R
e
sp
on
d R
a
nk
i
s
gi
ve
n
by
Resp
ond
Rank
(U
)
=
Resp
ond
Rank
(Jam
es)
=
Resp
ond
Rank
(Jam
es)
=
1
0
8
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE
Vo
l. 6
,
N
o
. 4
,
Au
gu
st 2
016
:
18
89
–
1
896
1
894
Tabl
e
3. R
e
s
p
o
ndR
a
n
k C
a
l
c
ul
at
i
o
n
Person Ja
m
e
s
Justine
T
o
tal follower
s
2000
500
Upvotes fr
o
m
follower
s
500
180
Upvotes fr
o
m
norm
a
l user
s
200
140
Answer
views fr
o
m
norm
a
l user
s
250
160
E
xper
t
upvotes
3
4
E
xper
t
downvotes
0
0
Special upvotes
0
0
RespondRa
n
k
108
127.
5
Also
t
h
e Respon
d
R
ank
o
f
Ju
st
in
e is calcu
lated
sim
ilarly
Resp
ond
Rank
(Ju
s
tin
e)
=
Resp
ond
Rank
(Ju
s
tin
e)
=127
.5
Here
Ju
stin
e go
t h
i
g
h
e
r
rank
th
an
t
h
e Jam
e
s and
he is ranked
top
.
W
ith this rank
ing
m
e
t
h
od
we can
calcu
late th
e ran
k
of an
y
n
u
m
b
er o
f
user
an
swer
p
a
r
ticip
atin
g
i
n
th
e
d
i
scu
ssi
on
and p
o
s
ition
th
e an
swers.
Th
e
Respo
n
d
R
an
k
is an
y
p
o
sitiv
e nu
m
b
er, i
f
th
e answer
co
n
t
ains m
o
re
dislik
es or
d
o
wn
v
o
t
es th
en
,
ran
k
is
negat
i
v
e
.
T
h
e
adva
nt
age
o
f
t
h
i
s
m
e
t
hod o
v
e
r exi
s
t
i
n
g a
p
p
r
oac
h
es i
s
t
h
i
s
ran
k
i
n
g m
e
t
hod t
r
eat
s al
l
t
h
e user
s
u
n
d
e
r th
e sam
e
criteria,
on
ly th
e
q
u
a
lity an
swer will b
e
in
t
h
e top
po
sitio
ns.
In
R
e
sp
o
ndR
a
n
k
m
e
t
hod an
answ
er i
s
onl
y
m
a
rked as t
o
p
answe
r
i
f
i
t
cont
ai
n
s
hi
g
h
p
e
rcent
a
ge of
fo
llo
wer up
vo
t
e
s
and als
o
it contains a
high
percent
age
of
a
n
sw
er vi
e
w
. These
are t
w
o features
we use
d
in
f
i
nd
ing
Resp
ond
Rank
and
some o
t
h
e
r
f
eatur
es lik
e th
e
size of the a
n
s
w
ers is not taken i
n
to consi
d
erati
on
because som
e
answer
with a
short
descri
ption will be
able
to satisfy t
h
e
user who as
ke
d the
question. Most
of
th
e
CQA
rank
i
n
g
is also
b
a
sed
on
th
e size of th
e an
swer
.
Bu
t so
m
e
u
s
er writes sm
al
ler q
u
a
lity an
swer th
at is
not
ra
n
k
e
d
t
o
p
due t
o
t
h
i
s
p
r
obl
em
. If t
h
e r
a
nk
of t
w
o
us
ers i
s
t
h
e sam
e
t
h
en,
we c
o
n
s
i
d
er t
h
e si
ze
of t
h
e
answ
er al
so f
o
r det
e
rm
i
n
i
ng t
h
e t
op ra
n
k
. S
o
m
e
ot
her fact
s consi
d
ere
d
i
f
t
h
e answe
r
i
n
g
of t
w
o
users i
s
t
h
e
sam
e
are the posting
date of t
h
e a
n
s
w
er, ef
fe
ctive use
of m
u
ltimedia in a
n
s
w
er etc.
5.
RESULTS
A
N
D
DI
SC
US
S
I
ON
We collected data-sets from
Quora, where t
h
e ave
r
ag
e turn
arou
nd
ti
m
e
o
f
th
e
user to
ge
t answers to
h
i
s q
u
e
stio
n
is f
o
ur
d
a
ys.
We co
m
p
u
t
ed
Respo
n
d
R
ank
fo
r
600
qu
estion
s
and
aro
und 9
0
0
0
an
sw
er
s. Th
e
gene
ral
st
at
i
s
t
i
c
s o
f
t
h
e
dat
a
-s
et
are
gi
ve
n i
n
Tabl
e
4.
Tab
l
e
4
.
Th
e gen
e
ral statistics of th
e d
a
taset
T
opics
Nu
m
b
e
r
of
questions
Nu
m
b
e
r
of
answers
Nu
m
b
er
of top
a
n
sw
e
r
diffe
r
fr
o
m
r
e
spondr
ank
#E
thnic and cultur
a
l differ
e
nces
50
456
6
#Philoso
phy
of ev
er
y
d
ay
life
87
1024
9
# Mini
m
a
list
lifestyle
12
86
3
# W
r
iter
s
and auth
or
s
53
436
7
#Star
bucks
46
502
5
#Snacks
23
412
2
#Health 86
1350
8
#Higher
education
68
956
8
#L
ife 79
1635
7
#Hu
m
an behavior
32
632
3
#W
or
king out
64
1236
7
T
o
tal
600
8725
65
M
o
st
of
past
r
e
search i
n
Q
u
ora i
s
f
o
r e
x
p
e
rt
i
d
ent
i
f
i
cat
i
on a
nd c
o
m
p
ari
s
i
on
wi
t
h
ot
her C
Q
A,
n
o
ran
k
i
n
g
i
s
pr
o
pos
ed f
o
r Q
u
o
r
a.S
o
we
c
o
m
p
are
d
ou
r
R
e
s
p
o
n
d
R
a
n
k
wi
t
h
e
x
i
s
t
i
ng Q
u
ora ra
nki
ng
a
n
d
t
h
e
ch
ang
e
s w
e
id
en
tif
ied
is
g
i
v
e
n
in
tab
l
e
5.Am
on
g t
h
ese 872
5
answers we ev
al
uat
e
d wi
t
h
Respon
dR
ank ,
65 t
o
p
answers
differ
from
existing top a
n
swers a
n
d
m
o
re
th
an
35
00
an
sw
er
s
p
o
si
t
i
on i
s
chan
ged.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Resp
on
d
Ra
nk:
Im
pr
ovi
n
g
Ra
nki
n
g
of
A
n
sw
e
rs i
n
C
o
m
m
uni
t
y
Que
s
t
i
o
n
An
sw
eri
n
g
(
G
eert
h
i
k
S)
1
895
Tab
l
e 5
.
C
o
m
p
ar
isio
n
of
r
e
spo
ndr
ank
w
ith
qu
or
a r
a
nk
T
o
tal answer
s
evaluated
8725
Nu
m
b
er
of changes in position with existing answer
s
3500
Nu
m
b
er
of changes in top position o
f
answer
s
65
W
e
also observed am
ong the
s
e 65 top ans
w
ers written
use
r
s, m
o
re than 50
users ha
d less than 100
fol
l
o
wers.
W
e
can see t
h
at
exi
s
t
i
ng ranki
n
g
i
s
i
n
favour o
f
users ha
vi
ng
great
er num
ber of f
o
l
l
o
wers
. Al
so
m
o
re
t
h
an 40
percent
of ans
w
er p
o
si
t
i
ons are change
d t
h
at
i
n
favo
ur
of users
havi
n
g
great
er num
ber of
fol
l
o
wers.T
hi
s resul
t
s
shows
t
h
at
our
R
e
sp
on
dR
ank
al
g
o
ri
t
h
m
i
d
ent
i
f
ies
m
o
st
co
m
p
rehensi
v
e, t
r
ust
w
ort
h
y
answers c
o
m
p
ared t
o
t
h
e exi
s
t
i
ng ranki
n
g
m
e
t
hods.
The ne
xt
st
age
of
ou
r resea
r
c
h
i
s
base
d o
n
t
h
e dat
a
-set
of
users
ha
vi
n
g
m
o
re t
h
an 2
0
0
0
f
o
l
l
o
we
rs
.
W
e
selected
25
user
s i
n
Q
uor
a
w
h
o
h
a
d
m
o
r
e
th
an
200
0 fo
llo
w
e
r
s
.
Th
e
an
alysis is d
one o
n
th
e
p
a
st an
sw
ers
written
b
y
th
em
. W
e
selected
1
0
0
d
i
fferen
t an
swers wr
itten
b
y
th
ese
25 u
s
ers an
d we
foun
d
t
h
at 68
an
swers
am
ong
10
0 an
swers a
r
e sel
ect
ed as t
op ans
w
ers
.
Fr
om
t
h
i
s
i
t
i
s
cl
ear t
h
at
i
f
t
h
e user ha
ve a l
a
rge n
u
m
b
er o
f
fo
llowers th
en
,
prob
ab
ility o
f
g
e
ttin
g
h
i
s answers as top
an
swer is
v
e
ry
h
i
gh
. Fro
m
a
b
ov
e two
resu
lts it is
clear
that our RespondRank m
e
thod
is
hel
p
ful for fi
nding
quality ans
w
ers written
by a
ny num
ber
of us
ers.
6.
CO
NCL
USI
O
N
AN
D F
U
T
U
RE W
O
R
K
This pa
pe
r lists the proble
m
faced by
users in
CQA si
tes
and
uses
a
RespondRa
n
k m
e
thod
for
rank
ing
an
swers. Th
e m
a
in
prob
lem
we id
en
tified
i
n
CQA
sites is
qu
ality an
swers written
b
y
n
e
w
u
s
ers are
not
pr
o
p
erl
y
re
cog
n
i
zed
.
In
o
u
r
R
e
sp
o
nd R
a
nk
m
e
t
hod a
b
o
v
e
pr
obl
em
are sol
v
e
d
by
i
n
t
r
o
d
u
ci
n
g
perce
n
t
a
g
e
of f
o
l
l
o
wer u
p
v
o
t
e
s, pe
rce
n
t
a
ge o
f
ans
w
er vi
ew as ne
w ran
k
i
n
g pa
r
a
m
e
t
e
rs. The expe
rt
u
p
vot
es an
d
downvotes are
also take
n int
o
consi
d
eration in
our
ra
nki
ng m
e
thod
which increa
ses the reliability of
our
ran
k
i
n
g m
e
t
hod. T
h
e
R
e
s
p
o
n
d
R
a
n
k
al
go
ri
t
h
m
i
s
eval
uat
e
d wi
t
h
Q
u
ora
dat
a
set
s
an
d
i
t
i
d
ent
i
f
i
e
s t
h
e best
answ
er
bet
t
e
r t
h
an
exi
s
t
i
n
g m
e
t
h
o
d
s.
I
n
ou
r
expe
ri
m
e
nt
s we ap
pl
i
e
d
R
e
s
p
on
dR
a
n
k
al
g
o
r
i
t
h
m
for
fi
ndi
n
g
t
h
e
b
e
st an
swer and
po
sition
i
ng
user
an
swers.
We can
m
o
d
i
fy
Resp
ond
Rank
algorith
m
fo
r rank
ing
th
e
qu
estion
s
in
CQA
als
o
. T
h
ere is a need
to id
entify the
best question
because som
e
e
xpe
rts only ans
w
er fe
we
r que
stions
d
u
e
to
t
h
eir ti
me av
ailab
ility
and
if th
e
C
Q
A
p
r
o
v
i
d
e
s
facilit
y to
id
en
tify b
e
st
qu
estio
ns, th
ey can
an
swer th
at
specific que
stion.
REFERE
NC
ES
[1]
A. Anderson,
et al.
, “Discovering value from community
activ
ity
on
focus
e
d CQA s
ites
:
a case s
t
ud
y
of s
t
ack
overflow,”
Proceedings of th
e 1
8
th ACM SIGKDD internati
ona
l conferen
ce on
Knowledge dis
c
overy and da
ta
mining. ACM
, 2
012.
[2]
C. Rughin
i
s,
et a
l
.
, “Computer-supported
collabor
ativ
e accoun
ts of
ma
jor d
e
pressio
n
: Digital rhetoric on Quor
a and
Wikipedia,”
20
14 9th Ib
erian Conferen
ce on Informat
ion Systems a
nd Technologies (
C
ISTI
)
, 2014.
doi:10.1109
/cisti.2014.6876968
.
[3]
R. Rughinis,
et al.
, “Computer-s
upported co
llab
o
rativ
e question
i
ng.
Regimes of
online sociality
on Quora,”
2
0
14
9th Iberian
Conference on In
forma
tion Systems and Technolog
ies (
C
ISTI
)
, 2014.
doi:10.1109
/cisti.2014.6876946
.
[4]
L. Bowler,
et al.
, “I know
what
y
ou are going thr
ough”: Answers
to
informational questions
about eating disorders
in Yahoo! answers: A qualitative stud
y
,
”
Pr
oceed
ings
of th
e Amer
ican Soc
iet
y
for
Infor
m
ation Sci
e
nc
e and
Technology
, vo
l/issue: 50(1), pp.
1–9,
2013
. doi:1
0.1002/meet.145
05001057.
[5]
X. Liu,
et al.
,
“ZhihuRank: A Topic-Sensitiv
e Expert Finding
Algorithm in
Com
m
unity
Qu
estion Answerin
g
We
bsi
t
e
s,
”
Lecture Notes
in Co
mputer Science
,
pp. 165–173
, 20
15. doi:10.1007
/978-3-319-25515-6_15.
[6]
J. Zhang,
et al.
,
“
E
xpertise net
w
orks in online com
m
unities: structur
e and alg
o
rithm
s
,”
Proceedings of the 16
th
internationa
l
co
nference on
World Wid
e
W
e
b. ACM
, 2007
.
http
://dx.doi.org/10
.
1
145/1242572.12
42603.
[7]
Z. M. Zhou,
et al.
, “Exploiting
user profile info
rmation for answer ranking in cQA,”
Proceedings of the 21st
International
Conference Companion on
World Wide Web - W
WW ’12
Companion
, 2012.
doi:10.1145
/218
7980.2188199.
[8]
A. Bal
t
adzh
iev
a
and G. Chrup
a
ł
a, “Question Quality
in Communi
ty
Question Ans
w
ering Forums,”
ACM SIGKDD
Explorations
Newsletter
, vol/issue: 17(1)
, pp
. 8–1
3, 2015
. doi:10.1
145/2830544.28
30547.
[9]
H. Toba,
et a
l
.
,
“Discovering high quality
answers
in community
question answ
ering
ar
chiv
es
us
ing a hier
arch
y
of
clas
s
i
fi
ers
,
”
Info
rmation Sciences
, vol. 261, pp. 1
01-115, 2014
. ht
tp://dx
.
doi
.org/1
0.1016/j
.
ins.201
3.10.030
.
[10]
Chua,
et al.
, “Measuring
the
effectiv
eness of
answers in Yahoo!
Answers,”
Onlin
e Information
R
evi
ew
,
vol/issue
:
39(1), pp
. 104-1
18, 2015
. http://dx.doi.org
/10.11
08/oir-10-2014-
0232.
[11]
H. Akkineni,
et al.
, “Online C
r
owds Opinion-Mining it to
A
n
aly
z
e Curren
t
Trend: A Review,”
Internationa
l
Journal of Electrical and
Computer
Eng
i
neer
ing
, vol/issue: 5(5),
2
015.
[12]
A.
E.
Kary
awati,
et a
l
.
, “
O
ntolo
g
y
-bas
ed W
h
y
-
Ques
tion Anal
ys
is
Using Lexico
-S
y
n
tactic Patter
ns,”
Int
e
rnation
a
l
Journal of Electrical and
Computer
Eng
i
neer
ing
,
vol/issue:
5(2), p
p
. 318
, 2015
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE
Vo
l. 6
,
N
o
. 4
,
Au
gu
st 2
016
:
18
89
–
1
896
1
896
[13]
H.
Wenwen,
et al.
, “Ranking
potential rep
l
y
-
provide
rs in co
m
m
unity
CQA s
y
stem
,”
Communications
, Chin
a
,
vol/issue:
10(10)
, pp
. 125-136
, 2
013. ht
tp://dx.do
i.org/10
.1109/cc.2013.6650325.
[14]
W.
Wa
ng,
et al.
,
“
I
m
p
roving question retr
iev
a
l in
com
m
unit
y
CQ
A with lab
e
l r
a
n
k
ing,”
Neural N
e
tworks (
I
JCNN)
,
The 2011 International
Join
t Co
nference on
. I
E
EE
, 2011
. h
ttp:
//d
x.doi.org
/10.110
9/ijcnn
.2011.603
3242.
[15]
X.
Tu,
et al.
, “Analogical r
eason
ing for an
s
w
er ranking in s
o
ci
al
CQA,”
IEEE In
t
e
llig
ent Syst
ems
, vol/issue: 27(5),
pp. 28-35
, 2012
. http://dx
.
doi.o
rg
/10.1109/mis.20
10.130.
Evaluation Warning : The document was created with Spire.PDF for Python.